2 * Copyright (c) 2009 Nicira Networks.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
22 #include <netinet/in.h>
25 #include "classifier.h"
27 #include "discovery.h"
30 #include "fail-open.h"
32 #include "mac-learning.h"
36 #include "ofp-print.h"
38 #include "openflow/nicira-ext.h"
39 #include "openflow/openflow.h"
40 #include "openflow/openflow-mgmt.h"
41 #include "openvswitch/datapath-protocol.h"
45 #include "poll-loop.h"
46 #include "port-array.h"
55 #include "vconn-ssl.h"
58 #define THIS_MODULE VLM_ofproto
68 TABLEID_CLASSIFIER = 1
72 struct netdev *netdev;
73 struct ofp_phy_port opp; /* In host byte order. */
76 static void ofport_free(struct ofport *);
77 static void hton_ofp_phy_port(struct ofp_phy_port *);
79 static int xlate_actions(const union ofp_action *in, size_t n_in,
80 const flow_t *flow, struct ofproto *ofproto,
81 const struct ofpbuf *packet,
82 struct odp_actions *out, tag_type *tags,
83 bool *may_setup_flow);
88 uint16_t idle_timeout; /* In seconds from time of last use. */
89 uint16_t hard_timeout; /* In seconds from time of creation. */
90 long long int used; /* Last-used time (0 if never used). */
91 long long int created; /* Creation time. */
92 uint64_t packet_count; /* Number of packets received. */
93 uint64_t byte_count; /* Number of bytes received. */
94 uint64_t accounted_bytes; /* Number of bytes passed to account_cb. */
95 uint8_t tcp_flags; /* Bitwise-OR of all TCP flags seen. */
96 uint8_t ip_tos; /* Last-seen IP type-of-service. */
97 tag_type tags; /* Tags (set only by hooks). */
99 /* If 'super' is non-NULL, this rule is a subrule, that is, it is an
100 * exact-match rule (having cr.wc.wildcards of 0) generated from the
101 * wildcard rule 'super'. In this case, 'list' is an element of the
104 * If 'super' is NULL, this rule is a super-rule, and 'list' is the head of
105 * a list of subrules. A super-rule with no wildcards (where
106 * cr.wc.wildcards is 0) will never have any subrules. */
112 * A subrule has no actions (it uses the super-rule's actions). */
114 union ofp_action *actions;
118 * A super-rule with wildcard fields never has ODP actions (since the
119 * datapath only supports exact-match flows). */
120 bool installed; /* Installed in datapath? */
121 bool may_install; /* True ordinarily; false if actions must
122 * be reassessed for every packet. */
124 union odp_action *odp_actions;
128 rule_is_hidden(const struct rule *rule)
130 /* Subrules are merely an implementation detail, so hide them from the
132 if (rule->super != NULL) {
136 /* Rules with priority higher than UINT16_MAX are set up by ofproto itself
137 * (e.g. by in-band control) and are intentionally hidden from the
139 if (rule->cr.priority > UINT16_MAX) {
146 static struct rule *rule_create(struct rule *super, const union ofp_action *,
147 size_t n_actions, uint16_t idle_timeout,
148 uint16_t hard_timeout);
149 static void rule_free(struct rule *);
150 static void rule_destroy(struct ofproto *, struct rule *);
151 static struct rule *rule_from_cls_rule(const struct cls_rule *);
152 static void rule_insert(struct ofproto *, struct rule *,
153 struct ofpbuf *packet, uint16_t in_port);
154 static void rule_remove(struct ofproto *, struct rule *);
155 static bool rule_make_actions(struct ofproto *, struct rule *,
156 const struct ofpbuf *packet);
157 static void rule_install(struct ofproto *, struct rule *,
158 struct rule *displaced_rule);
159 static void rule_uninstall(struct ofproto *, struct rule *);
160 static void rule_post_uninstall(struct ofproto *, struct rule *);
165 struct pktbuf *pktbuf;
169 struct rconn_packet_counter *packet_in_counter;
171 /* Number of OpenFlow messages queued as replies to OpenFlow requests, and
172 * the maximum number before we stop reading OpenFlow requests. */
173 #define OFCONN_REPLY_MAX 100
174 struct rconn_packet_counter *reply_counter;
177 static struct ofconn *ofconn_create(struct ofproto *, struct rconn *);
178 static void ofconn_destroy(struct ofconn *, struct ofproto *);
179 static void ofconn_run(struct ofconn *, struct ofproto *);
180 static void ofconn_wait(struct ofconn *);
181 static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
182 struct rconn_packet_counter *counter);
186 uint64_t datapath_id; /* Datapath ID. */
187 uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
188 uint64_t mgmt_id; /* Management channel identifier. */
189 char *manufacturer; /* Manufacturer. */
190 char *hardware; /* Hardware. */
191 char *software; /* Software version. */
192 char *serial; /* Serial number. */
196 struct netdev_monitor *netdev_monitor;
197 struct port_array ports; /* Index is ODP port nr; ofport->opp.port_no is
199 struct shash port_by_name;
203 struct switch_status *switch_status;
204 struct status_category *ss_cat;
205 struct in_band *in_band;
206 struct discovery *discovery;
207 struct fail_open *fail_open;
208 struct pinsched *miss_sched, *action_sched;
209 struct executer *executer;
210 struct netflow *netflow;
213 struct classifier cls;
214 bool need_revalidate;
215 long long int next_expiration;
216 struct tag_set revalidate_set;
218 /* OpenFlow connections. */
219 struct list all_conns;
220 struct ofconn *controller;
221 struct pvconn **listeners;
223 struct pvconn **snoops;
226 /* Hooks for ovs-vswitchd. */
227 const struct ofhooks *ofhooks;
230 /* Used by default ofhooks. */
231 struct mac_learning *ml;
234 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
236 static const struct ofhooks default_ofhooks;
238 static uint64_t pick_datapath_id(struct dpif *, uint64_t fallback_dpid);
239 static uint64_t pick_fallback_dpid(void);
240 static void send_packet_in_miss(struct ofpbuf *, void *ofproto);
241 static void send_packet_in_action(struct ofpbuf *, void *ofproto);
242 static void update_used(struct ofproto *);
243 static void update_stats(struct rule *, const struct odp_flow_stats *);
244 static void expire_rule(struct cls_rule *, void *ofproto);
245 static bool revalidate_rule(struct ofproto *p, struct rule *rule);
246 static void revalidate_cb(struct cls_rule *rule_, void *p_);
248 static void handle_odp_msg(struct ofproto *, struct ofpbuf *);
250 static void handle_openflow(struct ofconn *, struct ofproto *,
253 static void refresh_port_group(struct ofproto *, unsigned int group);
254 static void update_port(struct ofproto *, const char *devname);
255 static int init_ports(struct ofproto *);
256 static void reinit_ports(struct ofproto *);
259 ofproto_create(const char *datapath, const struct ofhooks *ofhooks, void *aux,
260 struct ofproto **ofprotop)
262 struct netdev_monitor *netdev_monitor;
263 struct odp_stats stats;
270 /* Connect to datapath and start listening for messages. */
271 error = dpif_open(datapath, &dpif);
273 VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
276 error = dpif_get_dp_stats(dpif, &stats);
278 VLOG_ERR("failed to obtain stats for datapath %s: %s",
279 datapath, strerror(error));
283 error = dpif_recv_set_mask(dpif, ODPL_MISS | ODPL_ACTION);
285 VLOG_ERR("failed to listen on datapath %s: %s",
286 datapath, strerror(error));
290 dpif_flow_flush(dpif);
291 dpif_recv_purge(dpif);
293 /* Arrange to monitor datapath ports for status changes. */
294 error = netdev_monitor_create(&netdev_monitor);
296 VLOG_ERR("failed to starting monitoring datapath %s: %s",
297 datapath, strerror(error));
302 /* Initialize settings. */
303 p = xcalloc(1, sizeof *p);
304 p->fallback_dpid = pick_fallback_dpid();
305 p->datapath_id = pick_datapath_id(dpif, p->fallback_dpid);
306 VLOG_INFO("using datapath ID %012"PRIx64, p->datapath_id);
307 p->manufacturer = xstrdup("Nicira Networks, Inc.");
308 p->hardware = xstrdup("Reference Implementation");
309 p->software = xstrdup(VERSION BUILDNR);
310 p->serial = xstrdup("None");
312 /* Initialize datapath. */
314 p->netdev_monitor = netdev_monitor;
315 port_array_init(&p->ports);
316 shash_init(&p->port_by_name);
317 p->max_ports = stats.max_ports;
319 /* Initialize submodules. */
320 p->switch_status = switch_status_create(p);
324 p->miss_sched = p->action_sched = NULL;
328 /* Initialize flow table. */
329 classifier_init(&p->cls);
330 p->need_revalidate = false;
331 p->next_expiration = time_msec() + 1000;
332 tag_set_init(&p->revalidate_set);
334 /* Initialize OpenFlow connections. */
335 list_init(&p->all_conns);
336 p->controller = ofconn_create(p, rconn_create(15, 15));
337 p->controller->pktbuf = pktbuf_create();
338 p->controller->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
344 /* Initialize hooks. */
346 p->ofhooks = ofhooks;
350 p->ofhooks = &default_ofhooks;
352 p->ml = mac_learning_create();
355 /* Register switch status category. */
356 p->ss_cat = switch_status_register(p->switch_status, "remote",
357 rconn_status_cb, p->controller->rconn);
360 error = init_ports(p);
371 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
373 uint64_t old_dpid = p->datapath_id;
374 p->datapath_id = (datapath_id
376 : pick_datapath_id(p->dpif, p->fallback_dpid));
377 if (p->datapath_id != old_dpid) {
378 VLOG_INFO("datapath ID changed to %012"PRIx64, p->datapath_id);
379 rconn_reconnect(p->controller->rconn);
384 ofproto_set_mgmt_id(struct ofproto *p, uint64_t mgmt_id)
386 p->mgmt_id = mgmt_id;
390 ofproto_set_probe_interval(struct ofproto *p, int probe_interval)
392 probe_interval = probe_interval ? MAX(probe_interval, 5) : 0;
393 rconn_set_probe_interval(p->controller->rconn, probe_interval);
395 int trigger_duration = probe_interval ? probe_interval * 3 : 15;
396 fail_open_set_trigger_duration(p->fail_open, trigger_duration);
401 ofproto_set_max_backoff(struct ofproto *p, int max_backoff)
403 rconn_set_max_backoff(p->controller->rconn, max_backoff);
407 ofproto_set_desc(struct ofproto *p,
408 const char *manufacturer, const char *hardware,
409 const char *software, const char *serial)
412 free(p->manufacturer);
413 p->manufacturer = xstrdup(manufacturer);
417 p->hardware = xstrdup(hardware);
421 p->software = xstrdup(software);
425 p->serial = xstrdup(serial);
430 ofproto_set_in_band(struct ofproto *p, bool in_band)
432 if (in_band != (p->in_band != NULL)) {
434 in_band_create(p, p->switch_status, p->controller->rconn,
438 ofproto_set_discovery(p, false, NULL, true);
439 in_band_destroy(p->in_band);
442 rconn_reconnect(p->controller->rconn);
448 ofproto_set_discovery(struct ofproto *p, bool discovery,
449 const char *re, bool update_resolv_conf)
451 if (discovery != (p->discovery != NULL)) {
453 int error = ofproto_set_in_band(p, true);
457 error = discovery_create(re, update_resolv_conf,
458 p->dpif, p->switch_status,
464 discovery_destroy(p->discovery);
467 rconn_disconnect(p->controller->rconn);
468 } else if (discovery) {
469 discovery_set_update_resolv_conf(p->discovery, update_resolv_conf);
470 return discovery_set_accept_controller_re(p->discovery, re);
476 ofproto_set_controller(struct ofproto *ofproto, const char *controller)
478 if (ofproto->discovery) {
480 } else if (controller) {
481 if (strcmp(rconn_get_name(ofproto->controller->rconn), controller)) {
482 return rconn_connect(ofproto->controller->rconn, controller);
487 rconn_disconnect(ofproto->controller->rconn);
493 set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
494 const struct svec *svec)
496 struct pvconn **pvconns = *pvconnsp;
497 size_t n_pvconns = *n_pvconnsp;
501 for (i = 0; i < n_pvconns; i++) {
502 pvconn_close(pvconns[i]);
506 pvconns = xmalloc(svec->n * sizeof *pvconns);
508 for (i = 0; i < svec->n; i++) {
509 const char *name = svec->names[i];
510 struct pvconn *pvconn;
513 error = pvconn_open(name, &pvconn);
515 pvconns[n_pvconns++] = pvconn;
517 VLOG_ERR("failed to listen on %s: %s", name, strerror(error));
525 *n_pvconnsp = n_pvconns;
531 ofproto_set_listeners(struct ofproto *ofproto, const struct svec *listeners)
533 return set_pvconns(&ofproto->listeners, &ofproto->n_listeners, listeners);
537 ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops)
539 return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops);
543 ofproto_set_netflow(struct ofproto *ofproto, const struct svec *collectors,
544 uint8_t engine_type, uint8_t engine_id, bool add_id_to_iface)
546 if (collectors && collectors->n) {
547 if (!ofproto->netflow) {
548 ofproto->netflow = netflow_create();
550 netflow_set_engine(ofproto->netflow, engine_type, engine_id,
552 return netflow_set_collectors(ofproto->netflow, collectors);
554 netflow_destroy(ofproto->netflow);
555 ofproto->netflow = NULL;
561 ofproto_set_failure(struct ofproto *ofproto, bool fail_open)
564 struct rconn *rconn = ofproto->controller->rconn;
565 int trigger_duration = rconn_get_probe_interval(rconn) * 3;
566 if (!ofproto->fail_open) {
567 ofproto->fail_open = fail_open_create(ofproto, trigger_duration,
568 ofproto->switch_status,
571 fail_open_set_trigger_duration(ofproto->fail_open,
575 fail_open_destroy(ofproto->fail_open);
576 ofproto->fail_open = NULL;
581 ofproto_set_rate_limit(struct ofproto *ofproto,
582 int rate_limit, int burst_limit)
584 if (rate_limit > 0) {
585 if (!ofproto->miss_sched) {
586 ofproto->miss_sched = pinsched_create(rate_limit, burst_limit,
587 ofproto->switch_status);
588 ofproto->action_sched = pinsched_create(rate_limit, burst_limit,
591 pinsched_set_limits(ofproto->miss_sched, rate_limit, burst_limit);
592 pinsched_set_limits(ofproto->action_sched,
593 rate_limit, burst_limit);
596 pinsched_destroy(ofproto->miss_sched);
597 ofproto->miss_sched = NULL;
598 pinsched_destroy(ofproto->action_sched);
599 ofproto->action_sched = NULL;
604 ofproto_set_stp(struct ofproto *ofproto UNUSED, bool enable_stp)
608 VLOG_WARN("STP is not yet implemented");
616 ofproto_set_remote_execution(struct ofproto *ofproto, const char *command_acl,
617 const char *command_dir)
620 if (!ofproto->executer) {
621 return executer_create(command_acl, command_dir,
624 executer_set_acl(ofproto->executer, command_acl, command_dir);
627 executer_destroy(ofproto->executer);
628 ofproto->executer = NULL;
634 ofproto_get_datapath_id(const struct ofproto *ofproto)
636 return ofproto->datapath_id;
640 ofproto_get_mgmt_id(const struct ofproto *ofproto)
642 return ofproto->mgmt_id;
646 ofproto_get_probe_interval(const struct ofproto *ofproto)
648 return rconn_get_probe_interval(ofproto->controller->rconn);
652 ofproto_get_max_backoff(const struct ofproto *ofproto)
654 return rconn_get_max_backoff(ofproto->controller->rconn);
658 ofproto_get_in_band(const struct ofproto *ofproto)
660 return ofproto->in_band != NULL;
664 ofproto_get_discovery(const struct ofproto *ofproto)
666 return ofproto->discovery != NULL;
670 ofproto_get_controller(const struct ofproto *ofproto)
672 return rconn_get_name(ofproto->controller->rconn);
676 ofproto_get_listeners(const struct ofproto *ofproto, struct svec *listeners)
680 for (i = 0; i < ofproto->n_listeners; i++) {
681 svec_add(listeners, pvconn_get_name(ofproto->listeners[i]));
686 ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
690 for (i = 0; i < ofproto->n_snoops; i++) {
691 svec_add(snoops, pvconn_get_name(ofproto->snoops[i]));
696 ofproto_destroy(struct ofproto *p)
698 struct ofconn *ofconn, *next_ofconn;
699 struct ofport *ofport;
700 unsigned int port_no;
707 ofproto_flush_flows(p);
708 classifier_destroy(&p->cls);
710 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
712 ofconn_destroy(ofconn, p);
716 netdev_monitor_destroy(p->netdev_monitor);
717 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
720 shash_destroy(&p->port_by_name);
722 switch_status_destroy(p->switch_status);
723 in_band_destroy(p->in_band);
724 discovery_destroy(p->discovery);
725 fail_open_destroy(p->fail_open);
726 pinsched_destroy(p->miss_sched);
727 pinsched_destroy(p->action_sched);
728 executer_destroy(p->executer);
729 netflow_destroy(p->netflow);
731 switch_status_unregister(p->ss_cat);
733 for (i = 0; i < p->n_listeners; i++) {
734 pvconn_close(p->listeners[i]);
738 for (i = 0; i < p->n_snoops; i++) {
739 pvconn_close(p->snoops[i]);
743 mac_learning_destroy(p->ml);
749 ofproto_run(struct ofproto *p)
751 int error = ofproto_run1(p);
753 error = ofproto_run2(p, false);
759 process_port_change(struct ofproto *ofproto, int error, char *devname)
761 if (error == ENOBUFS) {
762 reinit_ports(ofproto);
764 update_port(ofproto, devname);
770 ofproto_run1(struct ofproto *p)
772 struct ofconn *ofconn, *next_ofconn;
777 for (i = 0; i < 50; i++) {
781 error = dpif_recv(p->dpif, &buf);
783 if (error == ENODEV) {
784 /* Someone destroyed the datapath behind our back. The caller
785 * better destroy us and give up, because we're just going to
786 * spin from here on out. */
787 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
788 VLOG_ERR_RL(&rl, "%s: datapath was destroyed externally",
795 handle_odp_msg(p, buf);
798 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
799 process_port_change(p, error, devname);
801 while ((error = netdev_monitor_poll(p->netdev_monitor,
802 &devname)) != EAGAIN) {
803 process_port_change(p, error, devname);
807 in_band_run(p->in_band);
810 char *controller_name;
811 if (rconn_is_connectivity_questionable(p->controller->rconn)) {
812 discovery_question_connectivity(p->discovery);
814 if (discovery_run(p->discovery, &controller_name)) {
815 if (controller_name) {
816 rconn_connect(p->controller->rconn, controller_name);
818 rconn_disconnect(p->controller->rconn);
823 fail_open_run(p->fail_open);
825 pinsched_run(p->miss_sched, send_packet_in_miss, p);
826 pinsched_run(p->action_sched, send_packet_in_action, p);
828 executer_run(p->executer);
831 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
833 ofconn_run(ofconn, p);
836 for (i = 0; i < p->n_listeners; i++) {
840 retval = pvconn_accept(p->listeners[i], OFP_VERSION, &vconn);
842 ofconn_create(p, rconn_new_from_vconn("passive", vconn));
843 } else if (retval != EAGAIN) {
844 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
848 for (i = 0; i < p->n_snoops; i++) {
852 retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
854 rconn_add_monitor(p->controller->rconn, vconn);
855 } else if (retval != EAGAIN) {
856 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
860 if (time_msec() >= p->next_expiration) {
861 COVERAGE_INC(ofproto_expiration);
862 p->next_expiration = time_msec() + 1000;
865 classifier_for_each(&p->cls, CLS_INC_ALL, expire_rule, p);
867 /* Let the hook know that we're at a stable point: all outstanding data
868 * in existing flows has been accounted to the account_cb. Thus, the
869 * hook can now reasonably do operations that depend on having accurate
870 * flow volume accounting (currently, that's just bond rebalancing). */
871 if (p->ofhooks->account_checkpoint_cb) {
872 p->ofhooks->account_checkpoint_cb(p->aux);
877 netflow_run(p->netflow);
883 struct revalidate_cbdata {
884 struct ofproto *ofproto;
885 bool revalidate_all; /* Revalidate all exact-match rules? */
886 bool revalidate_subrules; /* Revalidate all exact-match subrules? */
887 struct tag_set revalidate_set; /* Set of tags to revalidate. */
891 ofproto_run2(struct ofproto *p, bool revalidate_all)
893 if (p->need_revalidate || revalidate_all
894 || !tag_set_is_empty(&p->revalidate_set)) {
895 struct revalidate_cbdata cbdata;
897 cbdata.revalidate_all = revalidate_all;
898 cbdata.revalidate_subrules = p->need_revalidate;
899 cbdata.revalidate_set = p->revalidate_set;
900 tag_set_init(&p->revalidate_set);
901 COVERAGE_INC(ofproto_revalidate);
902 classifier_for_each(&p->cls, CLS_INC_EXACT, revalidate_cb, &cbdata);
903 p->need_revalidate = false;
910 ofproto_wait(struct ofproto *p)
912 struct ofconn *ofconn;
915 dpif_recv_wait(p->dpif);
916 dpif_port_poll_wait(p->dpif);
917 netdev_monitor_poll_wait(p->netdev_monitor);
918 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
922 in_band_wait(p->in_band);
925 discovery_wait(p->discovery);
928 fail_open_wait(p->fail_open);
930 pinsched_wait(p->miss_sched);
931 pinsched_wait(p->action_sched);
933 executer_wait(p->executer);
935 if (!tag_set_is_empty(&p->revalidate_set)) {
936 poll_immediate_wake();
938 if (p->need_revalidate) {
939 /* Shouldn't happen, but if it does just go around again. */
940 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
941 poll_immediate_wake();
942 } else if (p->next_expiration != LLONG_MAX) {
943 poll_timer_wait(p->next_expiration - time_msec());
945 for (i = 0; i < p->n_listeners; i++) {
946 pvconn_wait(p->listeners[i]);
948 for (i = 0; i < p->n_snoops; i++) {
949 pvconn_wait(p->snoops[i]);
954 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
956 tag_set_add(&ofproto->revalidate_set, tag);
960 ofproto_get_revalidate_set(struct ofproto *ofproto)
962 return &ofproto->revalidate_set;
966 ofproto_is_alive(const struct ofproto *p)
968 return p->discovery || rconn_is_alive(p->controller->rconn);
972 ofproto_send_packet(struct ofproto *p, const flow_t *flow,
973 const union ofp_action *actions, size_t n_actions,
974 const struct ofpbuf *packet)
976 struct odp_actions odp_actions;
979 error = xlate_actions(actions, n_actions, flow, p, packet, &odp_actions,
985 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
987 dpif_execute(p->dpif, flow->in_port, odp_actions.actions,
988 odp_actions.n_actions, packet);
993 ofproto_add_flow(struct ofproto *p,
994 const flow_t *flow, uint32_t wildcards, unsigned int priority,
995 const union ofp_action *actions, size_t n_actions,
999 rule = rule_create(NULL, actions, n_actions,
1000 idle_timeout >= 0 ? idle_timeout : 5 /* XXX */, 0);
1001 cls_rule_from_flow(&rule->cr, flow, wildcards, priority);
1002 rule_insert(p, rule, NULL, 0);
1006 ofproto_delete_flow(struct ofproto *ofproto, const flow_t *flow,
1007 uint32_t wildcards, unsigned int priority)
1011 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1015 rule_remove(ofproto, rule);
1020 destroy_rule(struct cls_rule *rule_, void *ofproto_)
1022 struct rule *rule = rule_from_cls_rule(rule_);
1023 struct ofproto *ofproto = ofproto_;
1025 /* Mark the flow as not installed, even though it might really be
1026 * installed, so that rule_remove() doesn't bother trying to uninstall it.
1027 * There is no point in uninstalling it individually since we are about to
1028 * blow away all the flows with dpif_flow_flush(). */
1029 rule->installed = false;
1031 rule_remove(ofproto, rule);
1035 ofproto_flush_flows(struct ofproto *ofproto)
1037 COVERAGE_INC(ofproto_flush);
1038 classifier_for_each(&ofproto->cls, CLS_INC_ALL, destroy_rule, ofproto);
1039 dpif_flow_flush(ofproto->dpif);
1040 if (ofproto->in_band) {
1041 in_band_flushed(ofproto->in_band);
1043 if (ofproto->fail_open) {
1044 fail_open_flushed(ofproto->fail_open);
1049 reinit_ports(struct ofproto *p)
1051 struct svec devnames;
1052 struct ofport *ofport;
1053 unsigned int port_no;
1054 struct odp_port *odp_ports;
1058 svec_init(&devnames);
1059 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
1060 svec_add (&devnames, (char *) ofport->opp.name);
1062 dpif_port_list(p->dpif, &odp_ports, &n_odp_ports);
1063 for (i = 0; i < n_odp_ports; i++) {
1064 svec_add (&devnames, odp_ports[i].devname);
1068 svec_sort_unique(&devnames);
1069 for (i = 0; i < devnames.n; i++) {
1070 update_port(p, devnames.names[i]);
1072 svec_destroy(&devnames);
1076 refresh_port_group(struct ofproto *p, unsigned int group)
1080 struct ofport *port;
1081 unsigned int port_no;
1083 assert(group == DP_GROUP_ALL || group == DP_GROUP_FLOOD);
1085 ports = xmalloc(port_array_count(&p->ports) * sizeof *ports);
1087 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1088 if (group == DP_GROUP_ALL || !(port->opp.config & OFPPC_NO_FLOOD)) {
1089 ports[n_ports++] = port_no;
1092 dpif_port_group_set(p->dpif, group, ports, n_ports);
1097 refresh_port_groups(struct ofproto *p)
1099 refresh_port_group(p, DP_GROUP_FLOOD);
1100 refresh_port_group(p, DP_GROUP_ALL);
1103 static struct ofport *
1104 make_ofport(const struct odp_port *odp_port)
1106 enum netdev_flags flags;
1107 struct ofport *ofport;
1108 struct netdev *netdev;
1112 error = netdev_open(odp_port->devname, NETDEV_ETH_TYPE_NONE, &netdev);
1114 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1115 "cannot be opened (%s)",
1116 odp_port->devname, odp_port->port,
1117 odp_port->devname, strerror(error));
1121 ofport = xmalloc(sizeof *ofport);
1122 ofport->netdev = netdev;
1123 ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port);
1124 netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
1125 memcpy(ofport->opp.name, odp_port->devname,
1126 MIN(sizeof ofport->opp.name, sizeof odp_port->devname));
1127 ofport->opp.name[sizeof ofport->opp.name - 1] = '\0';
1129 netdev_get_flags(netdev, &flags);
1130 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1132 netdev_get_carrier(netdev, &carrier);
1133 ofport->opp.state = carrier ? 0 : OFPPS_LINK_DOWN;
1135 netdev_get_features(netdev,
1136 &ofport->opp.curr, &ofport->opp.advertised,
1137 &ofport->opp.supported, &ofport->opp.peer);
1142 ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port)
1144 if (port_array_get(&p->ports, odp_port->port)) {
1145 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1148 } else if (shash_find(&p->port_by_name, odp_port->devname)) {
1149 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1158 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1160 const struct ofp_phy_port *a = &a_->opp;
1161 const struct ofp_phy_port *b = &b_->opp;
1163 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1164 return (a->port_no == b->port_no
1165 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1166 && !strcmp((char *) a->name, (char *) b->name)
1167 && a->state == b->state
1168 && a->config == b->config
1169 && a->curr == b->curr
1170 && a->advertised == b->advertised
1171 && a->supported == b->supported
1172 && a->peer == b->peer);
1176 send_port_status(struct ofproto *p, const struct ofport *ofport,
1179 /* XXX Should limit the number of queued port status change messages. */
1180 struct ofconn *ofconn;
1181 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
1182 struct ofp_port_status *ops;
1185 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1186 ops->reason = reason;
1187 ops->desc = ofport->opp;
1188 hton_ofp_phy_port(&ops->desc);
1189 queue_tx(b, ofconn, NULL);
1191 if (p->ofhooks->port_changed_cb) {
1192 p->ofhooks->port_changed_cb(reason, &ofport->opp, p->aux);
1197 ofport_install(struct ofproto *p, struct ofport *ofport)
1199 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1200 port_array_set(&p->ports, ofp_port_to_odp_port(ofport->opp.port_no),
1202 shash_add(&p->port_by_name, (char *) ofport->opp.name, ofport);
1206 ofport_remove(struct ofproto *p, struct ofport *ofport)
1208 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1209 port_array_set(&p->ports, ofp_port_to_odp_port(ofport->opp.port_no), NULL);
1210 shash_delete(&p->port_by_name,
1211 shash_find(&p->port_by_name, (char *) ofport->opp.name));
1215 ofport_free(struct ofport *ofport)
1218 netdev_close(ofport->netdev);
1224 update_port(struct ofproto *p, const char *devname)
1226 struct odp_port odp_port;
1227 struct ofport *ofport;
1230 COVERAGE_INC(ofproto_update_port);
1231 ofport = shash_find_data(&p->port_by_name, devname);
1232 error = dpif_port_query_by_name(p->dpif, devname, &odp_port);
1236 if (!ofport_conflicts(p, &odp_port)) {
1237 ofport = make_ofport(&odp_port);
1239 ofport_install(p, ofport);
1240 send_port_status(p, ofport, OFPPR_ADD);
1244 /* Modified port. */
1245 struct ofport *new_ofport = make_ofport(&odp_port);
1250 new_ofport->opp.config &= OFPPC_PORT_DOWN;
1251 new_ofport->opp.config |= ofport->opp.config & ~OFPPC_PORT_DOWN;
1252 if (ofport_equal(ofport, new_ofport)) {
1253 /* False alarm--no change. */
1254 ofport_free(new_ofport);
1256 ofport_remove(p, ofport);
1257 ofport_install(p, new_ofport);
1258 ofport_free(ofport);
1259 send_port_status(p, new_ofport, OFPPR_MODIFY);
1262 } else if (error == ENOENT || error == ENODEV) {
1265 send_port_status(p, ofport, OFPPR_DELETE);
1266 ofport_remove(p, ofport);
1267 ofport_free(ofport);
1270 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1271 "%s", strerror(error));
1274 refresh_port_groups(p);
1278 init_ports(struct ofproto *p)
1280 struct odp_port *ports;
1285 error = dpif_port_list(p->dpif, &ports, &n_ports);
1290 for (i = 0; i < n_ports; i++) {
1291 const struct odp_port *odp_port = &ports[i];
1292 if (!ofport_conflicts(p, odp_port)) {
1293 struct ofport *ofport = make_ofport(odp_port);
1295 ofport_install(p, ofport);
1300 refresh_port_groups(p);
1304 static struct ofconn *
1305 ofconn_create(struct ofproto *p, struct rconn *rconn)
1307 struct ofconn *ofconn = xmalloc(sizeof *ofconn);
1308 list_push_back(&p->all_conns, &ofconn->node);
1309 ofconn->rconn = rconn;
1310 ofconn->pktbuf = NULL;
1311 ofconn->send_flow_exp = false;
1312 ofconn->miss_send_len = 0;
1313 ofconn->packet_in_counter = rconn_packet_counter_create ();
1314 ofconn->reply_counter = rconn_packet_counter_create ();
1319 ofconn_destroy(struct ofconn *ofconn, struct ofproto *p)
1322 executer_rconn_closing(p->executer, ofconn->rconn);
1325 list_remove(&ofconn->node);
1326 rconn_destroy(ofconn->rconn);
1327 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1328 rconn_packet_counter_destroy(ofconn->reply_counter);
1329 pktbuf_destroy(ofconn->pktbuf);
1334 ofconn_run(struct ofconn *ofconn, struct ofproto *p)
1338 rconn_run(ofconn->rconn);
1340 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1341 /* Limit the number of iterations to prevent other tasks from
1343 for (iteration = 0; iteration < 50; iteration++) {
1344 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1348 handle_openflow(ofconn, p, of_msg);
1349 ofpbuf_delete(of_msg);
1353 if (ofconn != p->controller && !rconn_is_alive(ofconn->rconn)) {
1354 ofconn_destroy(ofconn, p);
1359 ofconn_wait(struct ofconn *ofconn)
1361 rconn_run_wait(ofconn->rconn);
1362 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1363 rconn_recv_wait(ofconn->rconn);
1365 COVERAGE_INC(ofproto_ofconn_stuck);
1369 /* Caller is responsible for initializing the 'cr' member of the returned
1371 static struct rule *
1372 rule_create(struct rule *super,
1373 const union ofp_action *actions, size_t n_actions,
1374 uint16_t idle_timeout, uint16_t hard_timeout)
1376 struct rule *rule = xcalloc(1, sizeof *rule);
1377 rule->idle_timeout = idle_timeout;
1378 rule->hard_timeout = hard_timeout;
1379 rule->used = rule->created = time_msec();
1380 rule->super = super;
1382 list_push_back(&super->list, &rule->list);
1384 list_init(&rule->list);
1386 rule->n_actions = n_actions;
1387 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1391 static struct rule *
1392 rule_from_cls_rule(const struct cls_rule *cls_rule)
1394 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
1398 rule_free(struct rule *rule)
1400 free(rule->actions);
1401 free(rule->odp_actions);
1405 /* Destroys 'rule'. If 'rule' is a subrule, also removes it from its
1406 * super-rule's list of subrules. If 'rule' is a super-rule, also iterates
1407 * through all of its subrules and revalidates them, destroying any that no
1408 * longer has a super-rule (which is probably all of them).
1410 * Before calling this function, the caller must make have removed 'rule' from
1411 * the classifier. If 'rule' is an exact-match rule, the caller is also
1412 * responsible for ensuring that it has been uninstalled from the datapath. */
1414 rule_destroy(struct ofproto *ofproto, struct rule *rule)
1417 struct rule *subrule, *next;
1418 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
1419 revalidate_rule(ofproto, subrule);
1422 list_remove(&rule->list);
1428 rule_has_out_port(const struct rule *rule, uint16_t out_port)
1430 const union ofp_action *oa;
1431 struct actions_iterator i;
1433 if (out_port == htons(OFPP_NONE)) {
1436 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
1437 oa = actions_next(&i)) {
1438 if (oa->type == htons(OFPAT_OUTPUT) && oa->output.port == out_port) {
1445 /* Executes the actions indicated by 'rule' on 'packet', which is in flow
1446 * 'flow' and is considered to have arrived on ODP port 'in_port'.
1448 * The flow that 'packet' actually contains does not need to actually match
1449 * 'rule'; the actions in 'rule' will be applied to it either way. Likewise,
1450 * the packet and byte counters for 'rule' will be credited for the packet sent
1451 * out whether or not the packet actually matches 'rule'.
1453 * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow,
1454 * the caller must already have accurately composed ODP actions for it given
1455 * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if
1456 * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this
1457 * function will compose a set of ODP actions based on 'rule''s OpenFlow
1458 * actions and apply them to 'packet'. */
1460 rule_execute(struct ofproto *ofproto, struct rule *rule,
1461 struct ofpbuf *packet, const flow_t *flow)
1463 const union odp_action *actions;
1465 struct odp_actions a;
1467 /* Grab or compose the ODP actions.
1469 * The special case for an exact-match 'rule' where 'flow' is not the
1470 * rule's flow is important to avoid, e.g., sending a packet out its input
1471 * port simply because the ODP actions were composed for the wrong
1473 if (rule->cr.wc.wildcards || !flow_equal(flow, &rule->cr.flow)) {
1474 struct rule *super = rule->super ? rule->super : rule;
1475 if (xlate_actions(super->actions, super->n_actions, flow, ofproto,
1476 packet, &a, NULL, 0)) {
1479 actions = a.actions;
1480 n_actions = a.n_actions;
1482 actions = rule->odp_actions;
1483 n_actions = rule->n_odp_actions;
1486 /* Execute the ODP actions. */
1487 if (!dpif_execute(ofproto->dpif, flow->in_port,
1488 actions, n_actions, packet)) {
1489 struct odp_flow_stats stats;
1490 flow_extract_stats(flow, packet, &stats);
1491 update_stats(rule, &stats);
1492 rule->used = time_msec();
1497 rule_insert(struct ofproto *p, struct rule *rule, struct ofpbuf *packet,
1500 struct rule *displaced_rule;
1502 /* Insert the rule in the classifier. */
1503 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
1504 if (!rule->cr.wc.wildcards) {
1505 rule_make_actions(p, rule, packet);
1508 /* Send the packet and credit it to the rule. */
1511 flow_extract(packet, in_port, &flow);
1512 rule_execute(p, rule, packet, &flow);
1515 /* Install the rule in the datapath only after sending the packet, to
1516 * avoid packet reordering. */
1517 if (rule->cr.wc.wildcards) {
1518 COVERAGE_INC(ofproto_add_wc_flow);
1519 p->need_revalidate = true;
1521 rule_install(p, rule, displaced_rule);
1524 /* Free the rule that was displaced, if any. */
1525 if (displaced_rule) {
1526 rule_destroy(p, displaced_rule);
1530 static struct rule *
1531 rule_create_subrule(struct ofproto *ofproto, struct rule *rule,
1534 struct rule *subrule = rule_create(rule, NULL, 0,
1535 rule->idle_timeout, rule->hard_timeout);
1536 COVERAGE_INC(ofproto_subrule_create);
1537 cls_rule_from_flow(&subrule->cr, flow, 0,
1538 (rule->cr.priority <= UINT16_MAX ? UINT16_MAX
1539 : rule->cr.priority));
1540 classifier_insert_exact(&ofproto->cls, &subrule->cr);
1546 rule_remove(struct ofproto *ofproto, struct rule *rule)
1548 if (rule->cr.wc.wildcards) {
1549 COVERAGE_INC(ofproto_del_wc_flow);
1550 ofproto->need_revalidate = true;
1552 rule_uninstall(ofproto, rule);
1554 classifier_remove(&ofproto->cls, &rule->cr);
1555 rule_destroy(ofproto, rule);
1558 /* Returns true if the actions changed, false otherwise. */
1560 rule_make_actions(struct ofproto *p, struct rule *rule,
1561 const struct ofpbuf *packet)
1563 const struct rule *super;
1564 struct odp_actions a;
1567 assert(!rule->cr.wc.wildcards);
1569 super = rule->super ? rule->super : rule;
1571 xlate_actions(super->actions, super->n_actions, &rule->cr.flow, p,
1572 packet, &a, &rule->tags, &rule->may_install);
1574 actions_len = a.n_actions * sizeof *a.actions;
1575 if (rule->n_odp_actions != a.n_actions
1576 || memcmp(rule->odp_actions, a.actions, actions_len)) {
1577 COVERAGE_INC(ofproto_odp_unchanged);
1578 free(rule->odp_actions);
1579 rule->n_odp_actions = a.n_actions;
1580 rule->odp_actions = xmemdup(a.actions, actions_len);
1588 do_put_flow(struct ofproto *ofproto, struct rule *rule, int flags,
1589 struct odp_flow_put *put)
1591 memset(&put->flow.stats, 0, sizeof put->flow.stats);
1592 put->flow.key = rule->cr.flow;
1593 put->flow.actions = rule->odp_actions;
1594 put->flow.n_actions = rule->n_odp_actions;
1596 return dpif_flow_put(ofproto->dpif, put);
1600 rule_install(struct ofproto *p, struct rule *rule, struct rule *displaced_rule)
1602 assert(!rule->cr.wc.wildcards);
1604 if (rule->may_install) {
1605 struct odp_flow_put put;
1606 if (!do_put_flow(p, rule,
1607 ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS,
1609 rule->installed = true;
1610 if (displaced_rule) {
1611 update_stats(rule, &put.flow.stats);
1612 rule_post_uninstall(p, displaced_rule);
1615 } else if (displaced_rule) {
1616 rule_uninstall(p, displaced_rule);
1621 rule_reinstall(struct ofproto *ofproto, struct rule *rule)
1623 if (rule->installed) {
1624 struct odp_flow_put put;
1625 COVERAGE_INC(ofproto_dp_missed);
1626 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put);
1628 rule_install(ofproto, rule, NULL);
1633 rule_update_actions(struct ofproto *ofproto, struct rule *rule)
1635 bool actions_changed = rule_make_actions(ofproto, rule, NULL);
1636 if (rule->may_install) {
1637 if (rule->installed) {
1638 if (actions_changed) {
1639 /* XXX should really do rule_post_uninstall() for the *old* set
1640 * of actions, and distinguish the old stats from the new. */
1641 struct odp_flow_put put;
1642 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put);
1645 rule_install(ofproto, rule, NULL);
1648 rule_uninstall(ofproto, rule);
1653 rule_account(struct ofproto *ofproto, struct rule *rule, uint64_t extra_bytes)
1655 uint64_t total_bytes = rule->byte_count + extra_bytes;
1657 if (ofproto->ofhooks->account_flow_cb
1658 && total_bytes > rule->accounted_bytes)
1660 ofproto->ofhooks->account_flow_cb(
1661 &rule->cr.flow, rule->odp_actions, rule->n_odp_actions,
1662 total_bytes - rule->accounted_bytes, ofproto->aux);
1663 rule->accounted_bytes = total_bytes;
1668 rule_uninstall(struct ofproto *p, struct rule *rule)
1670 assert(!rule->cr.wc.wildcards);
1671 if (rule->installed) {
1672 struct odp_flow odp_flow;
1674 odp_flow.key = rule->cr.flow;
1675 odp_flow.actions = NULL;
1676 odp_flow.n_actions = 0;
1677 if (!dpif_flow_del(p->dpif, &odp_flow)) {
1678 update_stats(rule, &odp_flow.stats);
1680 rule->installed = false;
1682 rule_post_uninstall(p, rule);
1687 rule_post_uninstall(struct ofproto *ofproto, struct rule *rule)
1689 struct rule *super = rule->super;
1691 rule_account(ofproto, rule, 0);
1692 if (ofproto->netflow) {
1693 struct ofexpired expired;
1694 expired.flow = rule->cr.flow;
1695 expired.packet_count = rule->packet_count;
1696 expired.byte_count = rule->byte_count;
1697 expired.used = rule->used;
1698 expired.created = rule->created;
1699 expired.tcp_flags = rule->tcp_flags;
1700 expired.ip_tos = rule->ip_tos;
1701 netflow_expire(ofproto->netflow, &expired);
1704 super->packet_count += rule->packet_count;
1705 super->byte_count += rule->byte_count;
1706 super->tcp_flags |= rule->tcp_flags;
1707 if (rule->packet_count) {
1708 super->ip_tos = rule->ip_tos;
1712 /* Reset counters to prevent double counting if the rule ever gets
1714 rule->packet_count = 0;
1715 rule->byte_count = 0;
1716 rule->accounted_bytes = 0;
1717 rule->tcp_flags = 0;
1722 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
1723 struct rconn_packet_counter *counter)
1725 update_openflow_length(msg);
1726 if (rconn_send(ofconn->rconn, msg, counter)) {
1732 send_error(const struct ofconn *ofconn, const struct ofp_header *oh,
1733 int error, const void *data, size_t len)
1736 struct ofp_error_msg *oem;
1738 if (!(error >> 16)) {
1739 VLOG_WARN_RL(&rl, "not sending bad error code %d to controller",
1744 COVERAGE_INC(ofproto_error);
1745 oem = make_openflow_xid(len + sizeof *oem, OFPT_ERROR,
1746 oh ? oh->xid : 0, &buf);
1747 oem->type = htons((unsigned int) error >> 16);
1748 oem->code = htons(error & 0xffff);
1749 memcpy(oem->data, data, len);
1750 queue_tx(buf, ofconn, ofconn->reply_counter);
1754 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
1757 size_t oh_length = ntohs(oh->length);
1758 send_error(ofconn, oh, error, oh, MIN(oh_length, 64));
1762 hton_ofp_phy_port(struct ofp_phy_port *opp)
1764 opp->port_no = htons(opp->port_no);
1765 opp->config = htonl(opp->config);
1766 opp->state = htonl(opp->state);
1767 opp->curr = htonl(opp->curr);
1768 opp->advertised = htonl(opp->advertised);
1769 opp->supported = htonl(opp->supported);
1770 opp->peer = htonl(opp->peer);
1774 handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh)
1776 struct ofp_header *rq = oh;
1777 queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter);
1782 handle_features_request(struct ofproto *p, struct ofconn *ofconn,
1783 struct ofp_header *oh)
1785 struct ofp_switch_features *osf;
1787 unsigned int port_no;
1788 struct ofport *port;
1790 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
1791 osf->datapath_id = htonll(p->datapath_id);
1792 osf->n_buffers = htonl(pktbuf_capacity());
1794 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
1795 OFPC_PORT_STATS | OFPC_MULTI_PHY_TX);
1796 osf->actions = htonl((1u << OFPAT_OUTPUT) |
1797 (1u << OFPAT_SET_VLAN_VID) |
1798 (1u << OFPAT_SET_VLAN_PCP) |
1799 (1u << OFPAT_STRIP_VLAN) |
1800 (1u << OFPAT_SET_DL_SRC) |
1801 (1u << OFPAT_SET_DL_DST) |
1802 (1u << OFPAT_SET_NW_SRC) |
1803 (1u << OFPAT_SET_NW_DST) |
1804 (1u << OFPAT_SET_TP_SRC) |
1805 (1u << OFPAT_SET_TP_DST));
1807 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1808 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
1811 queue_tx(buf, ofconn, ofconn->reply_counter);
1816 handle_get_config_request(struct ofproto *p, struct ofconn *ofconn,
1817 struct ofp_header *oh)
1820 struct ofp_switch_config *osc;
1824 /* Figure out flags. */
1825 dpif_get_drop_frags(p->dpif, &drop_frags);
1826 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
1827 if (ofconn->send_flow_exp) {
1828 flags |= OFPC_SEND_FLOW_EXP;
1832 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
1833 osc->flags = htons(flags);
1834 osc->miss_send_len = htons(ofconn->miss_send_len);
1835 queue_tx(buf, ofconn, ofconn->reply_counter);
1841 handle_set_config(struct ofproto *p, struct ofconn *ofconn,
1842 struct ofp_switch_config *osc)
1847 error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc);
1851 flags = ntohs(osc->flags);
1853 ofconn->send_flow_exp = (flags & OFPC_SEND_FLOW_EXP) != 0;
1855 if (ofconn == p->controller) {
1856 switch (flags & OFPC_FRAG_MASK) {
1857 case OFPC_FRAG_NORMAL:
1858 dpif_set_drop_frags(p->dpif, false);
1860 case OFPC_FRAG_DROP:
1861 dpif_set_drop_frags(p->dpif, true);
1864 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
1870 if ((ntohs(osc->miss_send_len) != 0) != (ofconn->miss_send_len != 0)) {
1871 if (ntohs(osc->miss_send_len) != 0) {
1872 ofconn->pktbuf = pktbuf_create();
1874 pktbuf_destroy(ofconn->pktbuf);
1878 ofconn->miss_send_len = ntohs(osc->miss_send_len);
1884 add_output_group_action(struct odp_actions *actions, uint16_t group)
1886 odp_actions_add(actions, ODPAT_OUTPUT_GROUP)->output_group.group = group;
1890 add_controller_action(struct odp_actions *actions,
1891 const struct ofp_action_output *oao)
1893 union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER);
1894 a->controller.arg = oao->max_len ? ntohs(oao->max_len) : UINT32_MAX;
1897 struct action_xlate_ctx {
1899 const flow_t *flow; /* Flow to which these actions correspond. */
1900 int recurse; /* Recursion level, via xlate_table_action. */
1901 struct ofproto *ofproto;
1902 const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a
1903 * null pointer if we are revalidating
1904 * without a packet to refer to. */
1907 struct odp_actions *out; /* Datapath actions. */
1908 tag_type *tags; /* Tags associated with OFPP_NORMAL actions. */
1909 bool may_setup_flow; /* True ordinarily; false if the actions must
1910 * be reassessed for every packet. */
1913 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
1914 struct action_xlate_ctx *ctx);
1917 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
1919 const struct ofport *ofport = port_array_get(&ctx->ofproto->ports, port);
1920 if (!ofport || !(ofport->opp.config & OFPPC_NO_FWD)) {
1921 odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port;
1925 static struct rule *
1926 lookup_valid_rule(struct ofproto *ofproto, const flow_t *flow)
1929 rule = rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
1931 /* The rule we found might not be valid, since we could be in need of
1932 * revalidation. If it is not valid, don't return it. */
1935 && ofproto->need_revalidate
1936 && !revalidate_rule(ofproto, rule)) {
1937 COVERAGE_INC(ofproto_invalidated);
1945 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
1947 if (!ctx->recurse) {
1952 flow.in_port = in_port;
1954 rule = lookup_valid_rule(ctx->ofproto, &flow);
1961 do_xlate_actions(rule->actions, rule->n_actions, ctx);
1968 xlate_output_action(struct action_xlate_ctx *ctx,
1969 const struct ofp_action_output *oao)
1973 switch (ntohs(oao->port)) {
1975 add_output_action(ctx, ctx->flow->in_port);
1978 xlate_table_action(ctx, ctx->flow->in_port);
1981 if (!ctx->ofproto->ofhooks->normal_cb(ctx->flow, ctx->packet,
1982 ctx->out, ctx->tags,
1983 ctx->ofproto->aux)) {
1984 COVERAGE_INC(ofproto_uninstallable);
1985 ctx->may_setup_flow = false;
1989 add_output_group_action(ctx->out, DP_GROUP_FLOOD);
1992 add_output_group_action(ctx->out, DP_GROUP_ALL);
1994 case OFPP_CONTROLLER:
1995 add_controller_action(ctx->out, oao);
1998 add_output_action(ctx, ODPP_LOCAL);
2001 odp_port = ofp_port_to_odp_port(ntohs(oao->port));
2002 if (odp_port != ctx->flow->in_port) {
2003 add_output_action(ctx, odp_port);
2010 xlate_nicira_action(struct action_xlate_ctx *ctx,
2011 const struct nx_action_header *nah)
2013 const struct nx_action_resubmit *nar;
2014 int subtype = ntohs(nah->subtype);
2016 assert(nah->vendor == htonl(NX_VENDOR_ID));
2018 case NXAST_RESUBMIT:
2019 nar = (const struct nx_action_resubmit *) nah;
2020 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2024 VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype);
2030 do_xlate_actions(const union ofp_action *in, size_t n_in,
2031 struct action_xlate_ctx *ctx)
2033 struct actions_iterator iter;
2034 const union ofp_action *ia;
2035 const struct ofport *port;
2037 port = port_array_get(&ctx->ofproto->ports, ctx->flow->in_port);
2038 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2039 port->opp.config & (eth_addr_equals(ctx->flow->dl_dst, stp_eth_addr)
2040 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
2041 /* Drop this flow. */
2045 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2046 uint16_t type = ntohs(ia->type);
2047 union odp_action *oa;
2051 xlate_output_action(ctx, &ia->output);
2054 case OFPAT_SET_VLAN_VID:
2055 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_VID);
2056 oa->vlan_vid.vlan_vid = ia->vlan_vid.vlan_vid;
2059 case OFPAT_SET_VLAN_PCP:
2060 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_PCP);
2061 oa->vlan_pcp.vlan_pcp = ia->vlan_pcp.vlan_pcp;
2064 case OFPAT_STRIP_VLAN:
2065 odp_actions_add(ctx->out, ODPAT_STRIP_VLAN);
2068 case OFPAT_SET_DL_SRC:
2069 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC);
2070 memcpy(oa->dl_addr.dl_addr,
2071 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2074 case OFPAT_SET_DL_DST:
2075 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST);
2076 memcpy(oa->dl_addr.dl_addr,
2077 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2080 case OFPAT_SET_NW_SRC:
2081 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC);
2082 oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2085 case OFPAT_SET_TP_SRC:
2086 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC);
2087 oa->tp_port.tp_port = ia->tp_port.tp_port;
2091 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
2095 VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type);
2102 xlate_actions(const union ofp_action *in, size_t n_in,
2103 const flow_t *flow, struct ofproto *ofproto,
2104 const struct ofpbuf *packet,
2105 struct odp_actions *out, tag_type *tags, bool *may_setup_flow)
2107 tag_type no_tags = 0;
2108 struct action_xlate_ctx ctx;
2109 COVERAGE_INC(ofproto_ofp2odp);
2110 odp_actions_init(out);
2113 ctx.ofproto = ofproto;
2114 ctx.packet = packet;
2116 ctx.tags = tags ? tags : &no_tags;
2117 ctx.may_setup_flow = true;
2118 do_xlate_actions(in, n_in, &ctx);
2119 if (may_setup_flow) {
2120 *may_setup_flow = ctx.may_setup_flow;
2122 if (odp_actions_overflow(out)) {
2123 odp_actions_init(out);
2124 return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY);
2130 handle_packet_out(struct ofproto *p, struct ofconn *ofconn,
2131 struct ofp_header *oh)
2133 struct ofp_packet_out *opo;
2134 struct ofpbuf payload, *buffer;
2135 struct odp_actions actions;
2141 error = check_ofp_packet_out(oh, &payload, &n_actions, p->max_ports);
2145 opo = (struct ofp_packet_out *) oh;
2147 COVERAGE_INC(ofproto_packet_out);
2148 if (opo->buffer_id != htonl(UINT32_MAX)) {
2149 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
2159 flow_extract(&payload, ofp_port_to_odp_port(ntohs(opo->in_port)), &flow);
2160 error = xlate_actions((const union ofp_action *) opo->actions, n_actions,
2161 &flow, p, &payload, &actions, NULL, NULL);
2166 dpif_execute(p->dpif, flow.in_port, actions.actions, actions.n_actions,
2168 ofpbuf_delete(buffer);
2174 update_port_config(struct ofproto *p, struct ofport *port,
2175 uint32_t config, uint32_t mask)
2177 mask &= config ^ port->opp.config;
2178 if (mask & OFPPC_PORT_DOWN) {
2179 if (config & OFPPC_PORT_DOWN) {
2180 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2182 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2185 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD)
2186 if (mask & REVALIDATE_BITS) {
2187 COVERAGE_INC(ofproto_costly_flags);
2188 port->opp.config ^= mask & REVALIDATE_BITS;
2189 p->need_revalidate = true;
2191 #undef REVALIDATE_BITS
2192 if (mask & OFPPC_NO_FLOOD) {
2193 port->opp.config ^= OFPPC_NO_FLOOD;
2194 refresh_port_group(p, DP_GROUP_FLOOD);
2196 if (mask & OFPPC_NO_PACKET_IN) {
2197 port->opp.config ^= OFPPC_NO_PACKET_IN;
2202 handle_port_mod(struct ofproto *p, struct ofp_header *oh)
2204 const struct ofp_port_mod *opm;
2205 struct ofport *port;
2208 error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm);
2212 opm = (struct ofp_port_mod *) oh;
2214 port = port_array_get(&p->ports,
2215 ofp_port_to_odp_port(ntohs(opm->port_no)));
2217 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
2218 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
2219 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
2221 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
2222 if (opm->advertise) {
2223 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
2229 static struct ofpbuf *
2230 make_stats_reply(uint32_t xid, uint16_t type, size_t body_len)
2232 struct ofp_stats_reply *osr;
2235 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
2236 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
2238 osr->flags = htons(0);
2242 static struct ofpbuf *
2243 start_stats_reply(const struct ofp_stats_request *request, size_t body_len)
2245 return make_stats_reply(request->header.xid, request->type, body_len);
2249 append_stats_reply(size_t nbytes, struct ofconn *ofconn, struct ofpbuf **msgp)
2251 struct ofpbuf *msg = *msgp;
2252 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
2253 if (nbytes + msg->size > UINT16_MAX) {
2254 struct ofp_stats_reply *reply = msg->data;
2255 reply->flags = htons(OFPSF_REPLY_MORE);
2256 *msgp = make_stats_reply(reply->header.xid, reply->type, nbytes);
2257 queue_tx(msg, ofconn, ofconn->reply_counter);
2259 return ofpbuf_put_uninit(*msgp, nbytes);
2263 handle_desc_stats_request(struct ofproto *p, struct ofconn *ofconn,
2264 struct ofp_stats_request *request)
2266 struct ofp_desc_stats *ods;
2269 msg = start_stats_reply(request, sizeof *ods);
2270 ods = append_stats_reply(sizeof *ods, ofconn, &msg);
2271 strncpy(ods->mfr_desc, p->manufacturer, sizeof ods->mfr_desc);
2272 strncpy(ods->hw_desc, p->hardware, sizeof ods->hw_desc);
2273 strncpy(ods->sw_desc, p->software, sizeof ods->sw_desc);
2274 strncpy(ods->serial_num, p->serial, sizeof ods->serial_num);
2275 queue_tx(msg, ofconn, ofconn->reply_counter);
2281 count_subrules(struct cls_rule *cls_rule, void *n_subrules_)
2283 struct rule *rule = rule_from_cls_rule(cls_rule);
2284 int *n_subrules = n_subrules_;
2292 handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn,
2293 struct ofp_stats_request *request)
2295 struct ofp_table_stats *ots;
2297 struct odp_stats dpstats;
2298 int n_exact, n_subrules, n_wild;
2300 msg = start_stats_reply(request, sizeof *ots * 2);
2302 /* Count rules of various kinds. */
2304 classifier_for_each(&p->cls, CLS_INC_EXACT, count_subrules, &n_subrules);
2305 n_exact = classifier_count_exact(&p->cls) - n_subrules;
2306 n_wild = classifier_count(&p->cls) - classifier_count_exact(&p->cls);
2309 dpif_get_dp_stats(p->dpif, &dpstats);
2310 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2311 memset(ots, 0, sizeof *ots);
2312 ots->table_id = TABLEID_HASH;
2313 strcpy(ots->name, "hash");
2314 ots->wildcards = htonl(0);
2315 ots->max_entries = htonl(dpstats.max_capacity);
2316 ots->active_count = htonl(n_exact);
2317 ots->lookup_count = htonll(dpstats.n_frags + dpstats.n_hit +
2319 ots->matched_count = htonll(dpstats.n_hit); /* XXX */
2321 /* Classifier table. */
2322 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2323 memset(ots, 0, sizeof *ots);
2324 ots->table_id = TABLEID_CLASSIFIER;
2325 strcpy(ots->name, "classifier");
2326 ots->wildcards = htonl(OFPFW_ALL);
2327 ots->max_entries = htonl(65536);
2328 ots->active_count = htonl(n_wild);
2329 ots->lookup_count = htonll(0); /* XXX */
2330 ots->matched_count = htonll(0); /* XXX */
2332 queue_tx(msg, ofconn, ofconn->reply_counter);
2337 handle_port_stats_request(struct ofproto *p, struct ofconn *ofconn,
2338 struct ofp_stats_request *request)
2340 struct ofp_port_stats *ops;
2342 struct ofport *port;
2343 unsigned int port_no;
2345 msg = start_stats_reply(request, sizeof *ops * 16);
2346 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
2347 struct netdev_stats stats;
2349 /* Intentionally ignore return value, since errors will set 'stats' to
2350 * all-1s, which is correct for OpenFlow, and netdev_get_stats() will
2352 netdev_get_stats(port->netdev, &stats);
2354 ops = append_stats_reply(sizeof *ops, ofconn, &msg);
2355 ops->port_no = htons(odp_port_to_ofp_port(port_no));
2356 memset(ops->pad, 0, sizeof ops->pad);
2357 ops->rx_packets = htonll(stats.rx_packets);
2358 ops->tx_packets = htonll(stats.tx_packets);
2359 ops->rx_bytes = htonll(stats.rx_bytes);
2360 ops->tx_bytes = htonll(stats.tx_bytes);
2361 ops->rx_dropped = htonll(stats.rx_dropped);
2362 ops->tx_dropped = htonll(stats.tx_dropped);
2363 ops->rx_errors = htonll(stats.rx_errors);
2364 ops->tx_errors = htonll(stats.tx_errors);
2365 ops->rx_frame_err = htonll(stats.rx_frame_errors);
2366 ops->rx_over_err = htonll(stats.rx_over_errors);
2367 ops->rx_crc_err = htonll(stats.rx_crc_errors);
2368 ops->collisions = htonll(stats.collisions);
2371 queue_tx(msg, ofconn, ofconn->reply_counter);
2375 struct flow_stats_cbdata {
2376 struct ofproto *ofproto;
2377 struct ofconn *ofconn;
2383 query_stats(struct ofproto *p, struct rule *rule,
2384 uint64_t *packet_countp, uint64_t *byte_countp)
2386 uint64_t packet_count, byte_count;
2387 struct rule *subrule;
2388 struct odp_flow *odp_flows;
2391 n_odp_flows = rule->cr.wc.wildcards ? list_size(&rule->list) : 1;
2392 odp_flows = xcalloc(1, n_odp_flows * sizeof *odp_flows);
2393 if (rule->cr.wc.wildcards) {
2395 LIST_FOR_EACH (subrule, struct rule, list, &rule->list) {
2396 odp_flows[i++].key = subrule->cr.flow;
2399 odp_flows[0].key = rule->cr.flow;
2402 packet_count = rule->packet_count;
2403 byte_count = rule->byte_count;
2404 if (!dpif_flow_get_multiple(p->dpif, odp_flows, n_odp_flows)) {
2406 for (i = 0; i < n_odp_flows; i++) {
2407 struct odp_flow *odp_flow = &odp_flows[i];
2408 packet_count += odp_flow->stats.n_packets;
2409 byte_count += odp_flow->stats.n_bytes;
2414 *packet_countp = packet_count;
2415 *byte_countp = byte_count;
2419 flow_stats_cb(struct cls_rule *rule_, void *cbdata_)
2421 struct rule *rule = rule_from_cls_rule(rule_);
2422 struct flow_stats_cbdata *cbdata = cbdata_;
2423 struct ofp_flow_stats *ofs;
2424 uint64_t packet_count, byte_count;
2425 size_t act_len, len;
2427 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
2431 act_len = sizeof *rule->actions * rule->n_actions;
2432 len = offsetof(struct ofp_flow_stats, actions) + act_len;
2434 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2436 ofs = append_stats_reply(len, cbdata->ofconn, &cbdata->msg);
2437 ofs->length = htons(len);
2438 ofs->table_id = rule->cr.wc.wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH;
2440 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &ofs->match);
2441 ofs->duration = htonl((time_msec() - rule->created) / 1000);
2442 ofs->priority = htons(rule->cr.priority);
2443 ofs->idle_timeout = htons(rule->idle_timeout);
2444 ofs->hard_timeout = htons(rule->hard_timeout);
2445 memset(ofs->pad2, 0, sizeof ofs->pad2);
2446 ofs->packet_count = htonll(packet_count);
2447 ofs->byte_count = htonll(byte_count);
2448 memcpy(ofs->actions, rule->actions, act_len);
2452 table_id_to_include(uint8_t table_id)
2454 return (table_id == TABLEID_HASH ? CLS_INC_EXACT
2455 : table_id == TABLEID_CLASSIFIER ? CLS_INC_WILD
2456 : table_id == 0xff ? CLS_INC_ALL
2461 handle_flow_stats_request(struct ofproto *p, struct ofconn *ofconn,
2462 const struct ofp_stats_request *osr,
2465 struct ofp_flow_stats_request *fsr;
2466 struct flow_stats_cbdata cbdata;
2467 struct cls_rule target;
2469 if (arg_size != sizeof *fsr) {
2470 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2472 fsr = (struct ofp_flow_stats_request *) osr->body;
2474 COVERAGE_INC(ofproto_flows_req);
2476 cbdata.ofconn = ofconn;
2477 cbdata.out_port = fsr->out_port;
2478 cbdata.msg = start_stats_reply(osr, 1024);
2479 cls_rule_from_match(&target, &fsr->match, 0);
2480 classifier_for_each_match(&p->cls, &target,
2481 table_id_to_include(fsr->table_id),
2482 flow_stats_cb, &cbdata);
2483 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
2487 struct aggregate_stats_cbdata {
2488 struct ofproto *ofproto;
2490 uint64_t packet_count;
2491 uint64_t byte_count;
2496 aggregate_stats_cb(struct cls_rule *rule_, void *cbdata_)
2498 struct rule *rule = rule_from_cls_rule(rule_);
2499 struct aggregate_stats_cbdata *cbdata = cbdata_;
2500 uint64_t packet_count, byte_count;
2502 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
2506 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2508 cbdata->packet_count += packet_count;
2509 cbdata->byte_count += byte_count;
2514 handle_aggregate_stats_request(struct ofproto *p, struct ofconn *ofconn,
2515 const struct ofp_stats_request *osr,
2518 struct ofp_aggregate_stats_request *asr;
2519 struct ofp_aggregate_stats_reply *reply;
2520 struct aggregate_stats_cbdata cbdata;
2521 struct cls_rule target;
2524 if (arg_size != sizeof *asr) {
2525 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2527 asr = (struct ofp_aggregate_stats_request *) osr->body;
2529 COVERAGE_INC(ofproto_agg_request);
2531 cbdata.out_port = asr->out_port;
2532 cbdata.packet_count = 0;
2533 cbdata.byte_count = 0;
2535 cls_rule_from_match(&target, &asr->match, 0);
2536 classifier_for_each_match(&p->cls, &target,
2537 table_id_to_include(asr->table_id),
2538 aggregate_stats_cb, &cbdata);
2540 msg = start_stats_reply(osr, sizeof *reply);
2541 reply = append_stats_reply(sizeof *reply, ofconn, &msg);
2542 reply->flow_count = htonl(cbdata.n_flows);
2543 reply->packet_count = htonll(cbdata.packet_count);
2544 reply->byte_count = htonll(cbdata.byte_count);
2545 queue_tx(msg, ofconn, ofconn->reply_counter);
2550 handle_stats_request(struct ofproto *p, struct ofconn *ofconn,
2551 struct ofp_header *oh)
2553 struct ofp_stats_request *osr;
2557 error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr,
2562 osr = (struct ofp_stats_request *) oh;
2564 switch (ntohs(osr->type)) {
2566 return handle_desc_stats_request(p, ofconn, osr);
2569 return handle_flow_stats_request(p, ofconn, osr, arg_size);
2571 case OFPST_AGGREGATE:
2572 return handle_aggregate_stats_request(p, ofconn, osr, arg_size);
2575 return handle_table_stats_request(p, ofconn, osr);
2578 return handle_port_stats_request(p, ofconn, osr);
2581 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
2584 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
2588 static long long int
2589 msec_from_nsec(uint64_t sec, uint32_t nsec)
2591 return !sec ? 0 : sec * 1000 + nsec / 1000000;
2595 update_time(struct rule *rule, const struct odp_flow_stats *stats)
2597 long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec);
2598 if (used > rule->used) {
2604 update_stats(struct rule *rule, const struct odp_flow_stats *stats)
2606 update_time(rule, stats);
2607 rule->packet_count += stats->n_packets;
2608 rule->byte_count += stats->n_bytes;
2609 rule->tcp_flags |= stats->tcp_flags;
2610 if (stats->n_packets) {
2611 rule->ip_tos = stats->ip_tos;
2616 add_flow(struct ofproto *p, struct ofconn *ofconn,
2617 struct ofp_flow_mod *ofm, size_t n_actions)
2619 struct ofpbuf *packet;
2624 rule = rule_create(NULL, (const union ofp_action *) ofm->actions,
2625 n_actions, ntohs(ofm->idle_timeout),
2626 ntohs(ofm->hard_timeout));
2627 cls_rule_from_match(&rule->cr, &ofm->match, ntohs(ofm->priority));
2631 if (ofm->buffer_id != htonl(UINT32_MAX)) {
2632 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
2636 rule_insert(p, rule, packet, in_port);
2637 ofpbuf_delete(packet);
2642 modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm,
2643 size_t n_actions, uint16_t command, struct rule *rule)
2645 if (rule_is_hidden(rule)) {
2649 if (command == OFPFC_DELETE) {
2650 rule_remove(p, rule);
2652 size_t actions_len = n_actions * sizeof *rule->actions;
2654 if (n_actions == rule->n_actions
2655 && !memcmp(ofm->actions, rule->actions, actions_len))
2660 free(rule->actions);
2661 rule->actions = xmemdup(ofm->actions, actions_len);
2662 rule->n_actions = n_actions;
2664 if (rule->cr.wc.wildcards) {
2665 COVERAGE_INC(ofproto_mod_wc_flow);
2666 p->need_revalidate = true;
2668 rule_update_actions(p, rule);
2676 modify_flows_strict(struct ofproto *p, const struct ofp_flow_mod *ofm,
2677 size_t n_actions, uint16_t command)
2683 flow_from_match(&flow, &wildcards, &ofm->match);
2684 rule = rule_from_cls_rule(classifier_find_rule_exactly(
2685 &p->cls, &flow, wildcards,
2686 ntohs(ofm->priority)));
2689 if (command == OFPFC_DELETE
2690 && ofm->out_port != htons(OFPP_NONE)
2691 && !rule_has_out_port(rule, ofm->out_port)) {
2695 modify_flow(p, ofm, n_actions, command, rule);
2700 struct modify_flows_cbdata {
2701 struct ofproto *ofproto;
2702 const struct ofp_flow_mod *ofm;
2709 modify_flows_cb(struct cls_rule *rule_, void *cbdata_)
2711 struct rule *rule = rule_from_cls_rule(rule_);
2712 struct modify_flows_cbdata *cbdata = cbdata_;
2714 if (cbdata->out_port != htons(OFPP_NONE)
2715 && !rule_has_out_port(rule, cbdata->out_port)) {
2719 modify_flow(cbdata->ofproto, cbdata->ofm, cbdata->n_actions,
2720 cbdata->command, rule);
2724 modify_flows_loose(struct ofproto *p, const struct ofp_flow_mod *ofm,
2725 size_t n_actions, uint16_t command)
2727 struct modify_flows_cbdata cbdata;
2728 struct cls_rule target;
2732 cbdata.out_port = (command == OFPFC_DELETE ? ofm->out_port
2733 : htons(OFPP_NONE));
2734 cbdata.n_actions = n_actions;
2735 cbdata.command = command;
2737 cls_rule_from_match(&target, &ofm->match, 0);
2739 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
2740 modify_flows_cb, &cbdata);
2745 handle_flow_mod(struct ofproto *p, struct ofconn *ofconn,
2746 struct ofp_flow_mod *ofm)
2751 error = check_ofp_message_array(&ofm->header, OFPT_FLOW_MOD, sizeof *ofm,
2752 sizeof *ofm->actions, &n_actions);
2757 normalize_match(&ofm->match);
2758 if (!ofm->match.wildcards) {
2759 ofm->priority = htons(UINT16_MAX);
2762 error = validate_actions((const union ofp_action *) ofm->actions,
2763 n_actions, p->max_ports);
2768 switch (ntohs(ofm->command)) {
2770 return add_flow(p, ofconn, ofm, n_actions);
2773 return modify_flows_loose(p, ofm, n_actions, OFPFC_MODIFY);
2775 case OFPFC_MODIFY_STRICT:
2776 return modify_flows_strict(p, ofm, n_actions, OFPFC_MODIFY);
2779 return modify_flows_loose(p, ofm, n_actions, OFPFC_DELETE);
2781 case OFPFC_DELETE_STRICT:
2782 return modify_flows_strict(p, ofm, n_actions, OFPFC_DELETE);
2785 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
2790 send_capability_reply(struct ofproto *p, struct ofconn *ofconn, uint32_t xid)
2792 struct ofmp_capability_reply *ocr;
2794 char capabilities[] = "com.nicira.mgmt.manager=false\n";
2796 ocr = make_openflow_xid(sizeof(*ocr), OFPT_VENDOR, xid, &b);
2797 ocr->header.header.vendor = htonl(NX_VENDOR_ID);
2798 ocr->header.header.subtype = htonl(NXT_MGMT);
2799 ocr->header.type = htons(OFMPT_CAPABILITY_REPLY);
2801 ocr->format = htonl(OFMPCOF_SIMPLE);
2802 ocr->mgmt_id = htonll(p->mgmt_id);
2804 ofpbuf_put(b, capabilities, strlen(capabilities));
2806 queue_tx(b, ofconn, ofconn->reply_counter);
2810 handle_ofmp(struct ofproto *p, struct ofconn *ofconn,
2811 struct ofmp_header *ofmph)
2813 size_t msg_len = ntohs(ofmph->header.header.length);
2814 if (msg_len < sizeof(*ofmph)) {
2815 VLOG_WARN_RL(&rl, "dropping short managment message: %d\n", msg_len);
2816 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2819 if (ofmph->type == htons(OFMPT_CAPABILITY_REQUEST)) {
2820 struct ofmp_capability_request *ofmpcr;
2822 if (msg_len < sizeof(struct ofmp_capability_request)) {
2823 VLOG_WARN_RL(&rl, "dropping short capability request: %d\n",
2825 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2828 ofmpcr = (struct ofmp_capability_request *)ofmph;
2829 if (ofmpcr->format != htonl(OFMPCAF_SIMPLE)) {
2830 /* xxx Find a better type than bad subtype */
2831 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
2834 send_capability_reply(p, ofconn, ofmph->header.header.xid);
2837 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
2842 handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg)
2844 struct ofp_vendor_header *ovh = msg;
2845 struct nicira_header *nh;
2847 if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) {
2848 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2850 if (ovh->vendor != htonl(NX_VENDOR_ID)) {
2851 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
2853 if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) {
2854 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2858 switch (ntohl(nh->subtype)) {
2859 case NXT_STATUS_REQUEST:
2860 return switch_status_handle_request(p->switch_status, ofconn->rconn,
2863 case NXT_ACT_SET_CONFIG:
2864 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); /* XXX */
2866 case NXT_ACT_GET_CONFIG:
2867 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); /* XXX */
2869 case NXT_COMMAND_REQUEST:
2871 return executer_handle_request(p->executer, ofconn->rconn, msg);
2876 return handle_ofmp(p, ofconn, msg);
2879 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
2883 handle_openflow(struct ofconn *ofconn, struct ofproto *p,
2884 struct ofpbuf *ofp_msg)
2886 struct ofp_header *oh = ofp_msg->data;
2889 COVERAGE_INC(ofproto_recv_openflow);
2891 case OFPT_ECHO_REQUEST:
2892 error = handle_echo_request(ofconn, oh);
2895 case OFPT_ECHO_REPLY:
2899 case OFPT_FEATURES_REQUEST:
2900 error = handle_features_request(p, ofconn, oh);
2903 case OFPT_GET_CONFIG_REQUEST:
2904 error = handle_get_config_request(p, ofconn, oh);
2907 case OFPT_SET_CONFIG:
2908 error = handle_set_config(p, ofconn, ofp_msg->data);
2911 case OFPT_PACKET_OUT:
2912 error = handle_packet_out(p, ofconn, ofp_msg->data);
2916 error = handle_port_mod(p, oh);
2920 error = handle_flow_mod(p, ofconn, ofp_msg->data);
2923 case OFPT_STATS_REQUEST:
2924 error = handle_stats_request(p, ofconn, oh);
2928 error = handle_vendor(p, ofconn, ofp_msg->data);
2932 if (VLOG_IS_WARN_ENABLED()) {
2933 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
2934 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
2937 error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
2942 send_error_oh(ofconn, ofp_msg->data, error);
2947 handle_odp_msg(struct ofproto *p, struct ofpbuf *packet)
2949 struct odp_msg *msg = packet->data;
2950 uint16_t in_port = odp_port_to_ofp_port(msg->port);
2952 struct ofpbuf payload;
2955 /* Handle controller actions. */
2956 if (msg->type == _ODPL_ACTION_NR) {
2957 COVERAGE_INC(ofproto_ctlr_action);
2958 pinsched_send(p->action_sched, in_port, packet,
2959 send_packet_in_action, p);
2963 payload.data = msg + 1;
2964 payload.size = msg->length - sizeof *msg;
2965 flow_extract(&payload, msg->port, &flow);
2967 rule = lookup_valid_rule(p, &flow);
2969 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
2970 struct ofport *port = port_array_get(&p->ports, msg->port);
2972 if (port->opp.config & OFPPC_NO_PACKET_IN) {
2973 COVERAGE_INC(ofproto_no_packet_in);
2974 /* XXX install 'drop' flow entry */
2975 ofpbuf_delete(packet);
2979 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, msg->port);
2982 COVERAGE_INC(ofproto_packet_in);
2983 pinsched_send(p->miss_sched, in_port, packet, send_packet_in_miss, p);
2987 if (rule->cr.wc.wildcards) {
2988 rule = rule_create_subrule(p, rule, &flow);
2989 rule_make_actions(p, rule, packet);
2991 if (!rule->may_install) {
2992 /* The rule is not installable, that is, we need to process every
2993 * packet, so process the current packet and set its actions into
2995 rule_make_actions(p, rule, packet);
2997 /* XXX revalidate rule if it needs it */
3001 rule_execute(p, rule, &payload, &flow);
3002 rule_reinstall(p, rule);
3003 ofpbuf_delete(packet);
3007 revalidate_cb(struct cls_rule *sub_, void *cbdata_)
3009 struct rule *sub = rule_from_cls_rule(sub_);
3010 struct revalidate_cbdata *cbdata = cbdata_;
3012 if (cbdata->revalidate_all
3013 || (cbdata->revalidate_subrules && sub->super)
3014 || (tag_set_intersects(&cbdata->revalidate_set, sub->tags))) {
3015 revalidate_rule(cbdata->ofproto, sub);
3020 revalidate_rule(struct ofproto *p, struct rule *rule)
3022 const flow_t *flow = &rule->cr.flow;
3024 COVERAGE_INC(ofproto_revalidate_rule);
3027 super = rule_from_cls_rule(classifier_lookup_wild(&p->cls, flow));
3029 rule_remove(p, rule);
3031 } else if (super != rule->super) {
3032 COVERAGE_INC(ofproto_revalidate_moved);
3033 list_remove(&rule->list);
3034 list_push_back(&super->list, &rule->list);
3035 rule->super = super;
3036 rule->hard_timeout = super->hard_timeout;
3037 rule->idle_timeout = super->idle_timeout;
3038 rule->created = super->created;
3043 rule_update_actions(p, rule);
3047 static struct ofpbuf *
3048 compose_flow_exp(const struct rule *rule, long long int now, uint8_t reason)
3050 struct ofp_flow_expired *ofe;
3053 ofe = make_openflow(sizeof *ofe, OFPT_FLOW_EXPIRED, &buf);
3054 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &ofe->match);
3055 ofe->priority = htons(rule->cr.priority);
3056 ofe->reason = reason;
3057 ofe->duration = (now - rule->created) / 1000;
3058 ofe->packet_count = rule->packet_count;
3059 ofe->byte_count = rule->byte_count;
3065 send_flow_exp(struct ofproto *p, struct rule *rule,
3066 long long int now, uint8_t reason)
3068 struct ofconn *ofconn;
3069 struct ofconn *prev;
3072 /* We limit the maximum number of queued flow expirations it by accounting
3073 * them under the counter for replies. That works because preventing
3074 * OpenFlow requests from being processed also prevents new flows from
3075 * being added (and expiring). (It also prevents processing OpenFlow
3076 * requests that would not add new flows, so it is imperfect.) */
3079 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3080 if (ofconn->send_flow_exp && rconn_is_connected(ofconn->rconn)) {
3082 queue_tx(ofpbuf_clone(buf), prev, ofconn->reply_counter);
3084 buf = compose_flow_exp(rule, now, reason);
3090 queue_tx(buf, prev, ofconn->reply_counter);
3095 uninstall_idle_flow(struct ofproto *ofproto, struct rule *rule)
3097 assert(rule->installed);
3098 assert(!rule->cr.wc.wildcards);
3101 rule_remove(ofproto, rule);
3103 rule_uninstall(ofproto, rule);
3108 expire_rule(struct cls_rule *cls_rule, void *p_)
3110 struct ofproto *p = p_;
3111 struct rule *rule = rule_from_cls_rule(cls_rule);
3112 long long int hard_expire, idle_expire, expire, now;
3114 hard_expire = (rule->hard_timeout
3115 ? rule->created + rule->hard_timeout * 1000
3117 idle_expire = (rule->idle_timeout
3118 && (rule->super || list_is_empty(&rule->list))
3119 ? rule->used + rule->idle_timeout * 1000
3121 expire = MIN(hard_expire, idle_expire);
3122 if (expire == LLONG_MAX) {
3123 if (rule->installed && time_msec() >= rule->used + 5000) {
3124 uninstall_idle_flow(p, rule);
3131 if (rule->installed && now >= rule->used + 5000) {
3132 uninstall_idle_flow(p, rule);
3137 COVERAGE_INC(ofproto_expired);
3138 if (rule->cr.wc.wildcards) {
3139 /* Update stats. (This code will be a no-op if the rule expired
3140 * due to an idle timeout, because in that case the rule has no
3141 * subrules left.) */
3142 struct rule *subrule, *next;
3143 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
3144 rule_remove(p, subrule);
3148 send_flow_exp(p, rule, now,
3150 ? OFPER_HARD_TIMEOUT : OFPER_IDLE_TIMEOUT));
3151 rule_remove(p, rule);
3155 update_used(struct ofproto *p)
3157 struct odp_flow *flows;
3162 error = dpif_flow_list_all(p->dpif, &flows, &n_flows);
3167 for (i = 0; i < n_flows; i++) {
3168 struct odp_flow *f = &flows[i];
3171 rule = rule_from_cls_rule(
3172 classifier_find_rule_exactly(&p->cls, &f->key, 0, UINT16_MAX));
3173 if (!rule || !rule->installed) {
3174 COVERAGE_INC(ofproto_unexpected_rule);
3175 dpif_flow_del(p->dpif, f);
3179 update_time(rule, &f->stats);
3180 rule_account(p, rule, f->stats.n_bytes);
3186 do_send_packet_in(struct ofconn *ofconn, uint32_t buffer_id,
3187 const struct ofpbuf *packet, int send_len)
3189 struct ofp_packet_in *opi;
3190 struct ofpbuf payload, *buf;
3191 struct odp_msg *msg;
3194 payload.data = msg + 1;
3195 payload.size = msg->length - sizeof *msg;
3197 send_len = MIN(send_len, payload.size);
3198 buf = ofpbuf_new(sizeof *opi + send_len);
3199 opi = put_openflow_xid(offsetof(struct ofp_packet_in, data),
3200 OFPT_PACKET_IN, 0, buf);
3201 opi->buffer_id = htonl(buffer_id);
3202 opi->total_len = htons(payload.size);
3203 opi->in_port = htons(odp_port_to_ofp_port(msg->port));
3204 opi->reason = msg->type == _ODPL_ACTION_NR ? OFPR_ACTION : OFPR_NO_MATCH;
3205 ofpbuf_put(buf, payload.data, MIN(send_len, payload.size));
3206 update_openflow_length(buf);
3207 rconn_send_with_limit(ofconn->rconn, buf, ofconn->packet_in_counter, 100);
3211 send_packet_in_action(struct ofpbuf *packet, void *p_)
3213 struct ofproto *p = p_;
3214 struct ofconn *ofconn;
3215 struct odp_msg *msg;
3218 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3219 if (ofconn == p->controller || ofconn->miss_send_len) {
3220 do_send_packet_in(ofconn, UINT32_MAX, packet, msg->arg);
3223 ofpbuf_delete(packet);
3227 send_packet_in_miss(struct ofpbuf *packet, void *p_)
3229 struct ofproto *p = p_;
3230 struct ofconn *ofconn;
3231 struct ofpbuf payload;
3232 struct odp_msg *msg;
3235 payload.data = msg + 1;
3236 payload.size = msg->length - sizeof *msg;
3237 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3238 if (ofconn->miss_send_len) {
3239 uint32_t buffer_id = pktbuf_save(ofconn->pktbuf, &payload,
3241 int send_len = (buffer_id != UINT32_MAX ? ofconn->miss_send_len
3243 do_send_packet_in(ofconn, buffer_id, packet, send_len);
3246 ofpbuf_delete(packet);
3250 pick_datapath_id(struct dpif *dpif, uint64_t fallback_dpid)
3252 char local_name[IF_NAMESIZE];
3253 uint8_t ea[ETH_ADDR_LEN];
3256 error = dpif_port_get_name(dpif, ODPP_LOCAL,
3257 local_name, sizeof local_name);
3259 error = netdev_nodev_get_etheraddr(local_name, ea);
3261 return eth_addr_to_uint64(ea);
3263 VLOG_WARN("could not get MAC address for %s (%s)",
3264 local_name, strerror(error));
3267 return fallback_dpid;
3271 pick_fallback_dpid(void)
3273 uint8_t ea[ETH_ADDR_LEN];
3274 eth_addr_random(ea);
3275 ea[0] = 0x00; /* Set Nicira OUI. */
3278 return eth_addr_to_uint64(ea);
3282 default_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
3283 struct odp_actions *actions, tag_type *tags,
3286 struct ofproto *ofproto = ofproto_;
3289 /* Drop frames for reserved multicast addresses. */
3290 if (eth_addr_is_reserved(flow->dl_dst)) {
3294 /* Learn source MAC (but don't try to learn from revalidation). */
3295 if (packet != NULL) {
3296 tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
3299 /* The log messages here could actually be useful in debugging,
3300 * so keep the rate limit relatively high. */
3301 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
3302 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
3303 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
3304 ofproto_revalidate(ofproto, rev_tag);
3308 /* Determine output port. */
3309 out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags);
3311 add_output_group_action(actions, DP_GROUP_FLOOD);
3312 } else if (out_port != flow->in_port) {
3313 odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port;
3321 static const struct ofhooks default_ofhooks = {
3323 default_normal_ofhook_cb,